CN109242797B - Image denoising method, system and medium based on homogeneous and heterogeneous region fusion - Google Patents
Image denoising method, system and medium based on homogeneous and heterogeneous region fusion Download PDFInfo
- Publication number
- CN109242797B CN109242797B CN201811061739.8A CN201811061739A CN109242797B CN 109242797 B CN109242797 B CN 109242797B CN 201811061739 A CN201811061739 A CN 201811061739A CN 109242797 B CN109242797 B CN 109242797B
- Authority
- CN
- China
- Prior art keywords
- sub
- block
- image
- denoising
- homogeneous
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Expired - Fee Related
Links
- 238000000034 method Methods 0.000 title claims abstract description 38
- 230000004927 fusion Effects 0.000 title claims abstract description 16
- 238000004422 calculation algorithm Methods 0.000 claims abstract description 56
- 238000012545 processing Methods 0.000 claims description 19
- 238000001914 filtration Methods 0.000 description 9
- 238000010586 diagram Methods 0.000 description 8
- 238000013507 mapping Methods 0.000 description 8
- 230000009466 transformation Effects 0.000 description 7
- 238000012935 Averaging Methods 0.000 description 6
- 230000000694 effects Effects 0.000 description 4
- 238000000638 solvent extraction Methods 0.000 description 4
- 238000012800 visualization Methods 0.000 description 4
- 230000006872 improvement Effects 0.000 description 3
- 238000000513 principal component analysis Methods 0.000 description 3
- 230000000007 visual effect Effects 0.000 description 3
- 238000013459 approach Methods 0.000 description 2
- 230000000903 blocking effect Effects 0.000 description 2
- 230000000295 complement effect Effects 0.000 description 2
- 230000006870 function Effects 0.000 description 2
- 239000011159 matrix material Substances 0.000 description 2
- 238000012986 modification Methods 0.000 description 2
- 230000004048 modification Effects 0.000 description 2
- 230000008569 process Effects 0.000 description 2
- 238000011160 research Methods 0.000 description 2
- 238000005070 sampling Methods 0.000 description 2
- 238000012360 testing method Methods 0.000 description 2
- 230000003044 adaptive effect Effects 0.000 description 1
- 238000004458 analytical method Methods 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 230000005540 biological transmission Effects 0.000 description 1
- 230000015572 biosynthetic process Effects 0.000 description 1
- 238000004364 calculation method Methods 0.000 description 1
- 230000008859 change Effects 0.000 description 1
- 238000004590 computer program Methods 0.000 description 1
- 230000007547 defect Effects 0.000 description 1
- 238000013461 design Methods 0.000 description 1
- 230000001066 destructive effect Effects 0.000 description 1
- 238000011161 development Methods 0.000 description 1
- 238000002059 diagnostic imaging Methods 0.000 description 1
- 238000005315 distribution function Methods 0.000 description 1
- 238000005516 engineering process Methods 0.000 description 1
- 238000010191 image analysis Methods 0.000 description 1
- 238000005259 measurement Methods 0.000 description 1
- 238000003909 pattern recognition Methods 0.000 description 1
- 230000000704 physical effect Effects 0.000 description 1
- 238000004088 simulation Methods 0.000 description 1
- 230000001629 suppression Effects 0.000 description 1
- 238000012549 training Methods 0.000 description 1
- 239000013598 vector Substances 0.000 description 1
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T5/00—Image enhancement or restoration
- G06T5/70—Denoising; Smoothing
Landscapes
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Image Processing (AREA)
Abstract
The invention discloses an image denoising method, a system and a medium based on homogeneous and heterogeneous region fusion, which comprises the following steps: step (1): setting the sliding step length of a window, dividing an original image into a plurality of sub-blocks according to the set size of the window, and calculating the weight coefficient of each sub-block; step (2): carrying out denoising treatment on all sub-blocks divided from the original image by adopting an LPG-PCA algorithm; denoising all sub-blocks divided from the original image by adopting a three-dimensional block matching BM3D algorithm; and (3): classifying each subblock into a homogeneous region or a heterogeneous region according to the comparison of the weight coefficient and a set threshold; and correspondingly fusing the sub-blocks denoised by the two algorithms according to the region category and the weight coefficient of each sub-block to obtain a fused image, namely the final image.
Description
Technical Field
The invention relates to an image denoising method, system and medium based on homogeneous and heterogeneous region fusion.
Background
With the rapid development of image technology, images are widely applied in medical imaging, pattern recognition and the like. However, the image is inevitably interfered by noise during the formation and transmission processes. Noise in the image tends to interleave with the signal, causing details of the image itself, such as: contour boundaries, lines, etc. become blurred. Therefore, denoising processing is necessary to the image containing noise, which is convenient for higher-level image analysis and understanding. How to reasonably inhibit noise in an image, remove information with different requirements and strengthen useful information in the image is achieved, so that target distinguishing or object interpretation is facilitated, and the problem of image denoising is mainly researched.
In practice, the noise is processed according to the digital characteristics of the noise and the difference between the gray value of the noise and the gray value of the surrounding signal, and the denoising process can be completed in an image space domain or an image transformation domain. The image space domain denoising is to perform data operation on an original image and directly process the gray value of a pixel. At present, image space domain denoising methods include mean filtering, median filtering, low-pass filtering and the like, and all the methods have a common characteristic that each pixel in an image is processed in the same mode without considering the characteristics of each pixel, so that the method is very effective in removing additional random noise, but the noise is removed, meanwhile, the image is seriously blurred, and particularly, the blurring is serious at the edge and the detail of the image. Another very effective image denoising method is based on image transform domain denoising, and its basic idea is: firstly, carrying out certain transformation on a noise-containing image, converting the noise-containing image from a spatial domain to a transformation domain, then processing a transformation coefficient in the transformation domain, then carrying out inverse transformation, and converting the noise-containing image from the transformation domain to an original spatial domain again, and finally realizing effective denoising.
However, due to the difference of image features, the strong variability of physical properties and data of images among images, the rich texture structure of some images, and the large number of homogeneous parts of some images, the current denoising algorithm, whether based on a spatial domain or a transform domain, is based on a certain simplified image model, and its denoising performance is only outstanding in a single aspect, so that it is impossible to apply a single denoising model to images with different features to denoise and obtain excellent effects. Therefore, the method has important significance for targeted denoising after the images are classified according to the characteristics. In recent years, many advanced denoising algorithms have appeared, such as three-dimensional Block Matching algorithm (Block Matching 3D, BM3D), Non-Local homomorphic Sparse Code (LSSC) and Non-Local Fast adaptive SAR image denoising algorithm (FANS), Principal Component Analysis (LPG-PCA) for Local Pixel Block Grouping, Non-Local mean (NLM) denoising algorithm, dictionary learning algorithm for Sparse representation (K-Means-discrete Value denoising, K-D), etc., which have strong denoising capability but weak detail protection on images, and have weak denoising capability but good texture protection on images. At present, no method has strong denoising capability and can well protect the details of an original image without generating artifact information.
Disclosure of Invention
In order to solve the defects of the prior art, the invention provides an image denoising method, a system and a medium based on homogeneous and heterogeneous region fusion; the algorithm denoising performance index peak signal-to-noise ratio and the structural similarity value are improved compared with a single denoising algorithm, and the visual effect and the detail protection are also superior to the single denoising algorithm.
In order to solve the technical problems, the invention adopts the following technical scheme:
as a first aspect of the invention, an image denoising method based on homogeneous and heterogeneous region fusion is provided;
the image denoising method based on homogeneous and heterogeneous region fusion comprises the following steps:
step (1): setting the sliding step length of a window, dividing an original image into a plurality of sub-blocks according to the set size of the window, and calculating the weight coefficient of each sub-block;
step (2): carrying out denoising treatment on all sub-blocks divided from the original image by adopting an LPG-PCA algorithm;
denoising all sub-blocks divided from the original image by adopting a three-dimensional block matching BM3D algorithm;
and (3): classifying each subblock into a homogeneous region or a heterogeneous region according to the comparison of the weight coefficient and a set threshold; and correspondingly fusing the sub-blocks denoised by the two algorithms according to the region category and the weight coefficient of each sub-block to obtain a fused image, namely the final image.
Further, the homogeneous region refers to: the gray values of all the pixel points in the area are within a set range.
Further, the heterogeneous region refers to: regions other than the homogeneous region.
Further, in the step (1), the step of calculating the weight coefficient of each sub-block is:
a step (101): calculating generalized likelihood ratio lambda of jth sub-blockj(x) Comprises the following steps:
wherein G represents the geometric mean of the jth sub-block, A represents the arithmetic mean of the jth sub-block,
a step (102): according to a generalized likelihood ratio λj(x) Calculating the weight omega (lambda) of the jth sub-blockj)：
Wherein the parameter lambda0Taking the intermediate value of the generalized likelihood ratio lambda (x) of all the sub-blocks, and taking the slope alpha as a set value.
Further, the step (3) comprises the following steps:
judging whether the jth sub-block of the image belongs to a homogeneous region or a heterogeneous region;
if the weight coefficient ω (λ)j) If the sub-block is greater than or equal to 0.5, the sub-block belongs to a heterogeneous region; for the sub-blocks belonging to the heterogeneous region, multiplying the pixel gray value of the jth sub-block subjected to denoising processing by the three-dimensional block matching BM3D algorithm by the weight omega (lambda)j) Then, multiplying the pixel gray value of the jth sub-block subjected to denoising processing by the LPG-PCA algorithm by (1-omega (lambda)j) Post-summation to obtain the pixel gray value of the jth sub-block of the fused image; at this time, the three-dimensional block matching BM3D algorithm slightly contributes to the sub-block, so that the detail information of the heterogeneous region can be better protected.
If the weight coefficient ω (λ)j) Less than 0.5, the sub-block belongs to the homogeneous region; for the sub-blocks belonging to the homogeneous region, multiplying the pixel gray value of the jth sub-block subjected to denoising processing by the three-dimensional block matching BM3D algorithm by the weight omega (lambda)j) Then, multiplying the pixel gray value of the jth sub-block subjected to denoising processing by the LPG-PCA algorithm by (1-omega (lambda)j) Post-summation to obtain a fused imageThe pixel gray value of the jth sub-block. At the moment, the LPG-PCA algorithm slightly contributes to the sub-block, so that a homogeneous region can be smoothed better, and the purpose of denoising is achieved.
And further obtaining pixel gray values of all sub-blocks of the fused image, namely obtaining the image of the original image after denoising.
The noun explains:
LPG-PCA: principal Component Analysis With Local Pixel Grouping (LPG-PCA). The algorithm expresses pixels to be processed in a training set and neighborhoods thereof into subblock vectors, groups the subblocks by utilizing block similarity measurement to obtain a sample matrix, centralizes the sample matrix, and then performs denoising processing by adopting principal component analysis.
BM 3D: three-dimensional Block Matching algorithm (Block Matching 3D, BM 3D). The algorithm combines a non-local filtering method with wavelet shrinkage and wiener filtering. Firstly, a three-dimensional stack is formed by each subblock to be processed and similar subblocks thereof by utilizing a block matching algorithm, then filtering is carried out in a three-dimensional wavelet transform domain by adopting a hard threshold value, and the filtered subblocks are placed back to the position in an original image. And further processing the filtered image by adopting a block matching algorithm and wiener filtering to obtain a final de-noised image.
Furthermore, the weighted fusion is carried out on the images subjected to the noise removal of the LPG-PCA and the BM3D according to the homogeneous region and the heterogeneous region, so that the noise is effectively reduced in the homogeneous region, and the generation of artificial artifacts is avoided; texture and detail of the image are better protected in heterogeneous areas.
As a second aspect of the present invention, an image denoising system based on homogeneous and heterogeneous region fusion is proposed;
image denoising system based on homogeneous and heterogeneous region fusion, including: the computer program product comprises a memory, a processor, and computer instructions stored on the memory and executed on the processor, wherein the computer instructions, when executed by the processor, perform the steps of any of the above methods.
As a third aspect of the present invention, a computer-readable storage medium is proposed;
a computer readable storage medium having stored thereon computer instructions which, when executed by a processor, perform the steps of any of the above methods.
Compared with the prior art, the invention has the beneficial effects that:
a soft classification strategy is provided, and good compromise is made between the aspects of detail protection and denoising. The algorithm is based on two existing advanced denoising basic tools, and the key point of the algorithm is to select two excellent denoising algorithms with complementary characteristics, and the key point is the design of a combined program. The invention adopts a soft combination mode, combines the weight coefficient changing from 0 to 1 to linearly output the linear combination of the two algorithms, and optimizes the denoising effect of the current single denoising algorithm. The denoising and classification of the images are a hot problem in the current research, and have important research significance. Therefore, the image denoising algorithm based on soft classification researched by the invention has profound significance in common image denoising application of a non-local method.
Drawings
The accompanying drawings, which are incorporated in and constitute a part of this application, illustrate embodiments of the application and, together with the description, serve to explain the application and are not intended to limit the application.
FIG. 1 is a flow chart of the present invention;
2(a) -2 (h) are comparison diagrams of different windows selected from Lena images for soft classification;
fig. 3(a) -3 (h) are comparison diagrams of Lena images containing noise and denoised by different methods.
Detailed Description
It should be noted that the following detailed description is exemplary and is intended to provide further explanation of the disclosure. Unless defined otherwise, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this application belongs.
It is noted that the terminology used herein is for the purpose of describing particular embodiments only and is not intended to be limiting of example embodiments according to the present application. As used herein, the singular forms "a", "an" and "the" are intended to include the plural forms as well, and it should be understood that when the terms "comprises" and/or "comprising" are used in this specification, they specify the presence of stated features, steps, operations, devices, components, and/or combinations thereof, unless the context clearly indicates otherwise.
As shown in fig. 1, the image denoising method based on homogeneous and heterogeneous region fusion specifically includes the following steps:
the method comprises the following steps: carrying out soft classification on the image according to textures, and dividing the image into a homogeneous region and a heterogeneous region;
step two: denoising by using the current advanced image denoising algorithm and respectively adopting different methods;
step three: the de-noised images obtained by different methods are effectively fused, so that the noise levels of different areas are effectively reduced, and meanwhile, the texture and the details of the images are protected.
The specific steps of the first step are as follows:
step (1-1): in order to classify the images according to the homogeneous region and the heterogeneous region, the ratio of the arithmetic mean A and the geometric mean G of the gray value images is used as statistical information for distinguishing the homogeneous region and the heterogeneous region. By a suitably large window calculation centered on the target pixel, the destructive effect of noise is reduced. To provide a full resolution dense area, we have the window slide across the image.
Step (1-2-1): increasing the size of the window for calculating the a/G ratio may reduce the variance of the estimated values, but may result in estimation errors. This is because the pixels entering the window have different properties after the window is enlarged. To solve this problem, we inspire a non-local denoising approach. To include more pixels, a relatively large search area is used to select those pixels that are more likely to coincide with the target pixel using the block similarity metric. Specifically, we find the K sub-blocks that are most similar to the sub-blocks to be processed using the block similarity metric over a large search range. The K sub-blocks form a three-dimensional stack. And then averaging the three-dimensional stack along the third dimension, and calculating the A/G statistics from the obtained average sub-blocks. This amounts to the use of a multi-visualization step in terms of improving data quality, which, although a relatively crude strategy, is a significant improvement in reliability.
Step (1-2-2): under some simplifying assumptions, A/G is the solution of the Generalized Likelihood Ratio (GLR) test, so we focus on the Ratio of A to G. For an image block of N pixels centered on a given target pixel, we propose two assumptions:
H0a homogeneous region: the signal amplitudes in the image blocks are equal;
H1a heterogeneous region: the signal amplitudes in the image blocks are unequal.
The corresponding GLR is:
wherein X is the pixel value, Xi|uiCompliance
Distribution of (2). Will be provided with
Substituting Λ (x) and calculating sup, we finally obtain a Generalized Likelihood Ratio (GLR) statistical expression, the logarithmic form of which is:
wherein G represents the geometric mean of the jth sub-block, A represents the arithmetic mean of the jth sub-block,
step (1-2-3): the quality of the weight mapping graph is improved by establishing the three-dimensional stack in the step (1-2-1) similar to virtual visualization processing. Reliable statistics are provided even with a relatively small estimation window, thereby ensuring high resolution. However, using a fixed size sub-block window results in discontinuities at the window boundaries. This phenomenon is due to the fact that each pixel belongs to the estimation window of its many neighbors, affecting their heterogeneity index. For example, strong noise on a uniform background will cause pixels within the entire window to be marked as heterogeneous regions. Finally, the pixels of the neighborhood that should belong to a homogeneous region are denoised as a heterogeneous region, producing artifacts in the output. To solve this problem, we use a simple heuristic strategy of computing weights based on sub-block size, which may include larger weights and smaller weights, and then averaging the sub-blocks of different sizes. This can reduce some unreasonable estimates and avoid the occurrence of artifacts from the artificial filtering.
Step (1-3): theoretically, we only need to set a threshold for reasonable classification decision. Although the λ (x) probability distribution function is not shown, it is known from the gamma distribution that when N is a large value (N >10), we approximate that its shape and scale parameters depend only on N and L. Thus, a desired false alarm probability can be set and the corresponding thresholds can be calculated and analyzed for any values of N and L. However, these theoretical results do not really solve our real image classification problem, because when the noise is strong, all classification decisions become unreliable unless we incorporate a large number of samples into the test. Therefore, the key issue is the size N (including the number of pixels) of the sampling window, which is related to whether we can obtain reliable resolution. In a small window, even in homogeneous regions, decision statistics are very difficult; on the other hand, if N is increased, the analysis window may easily contain pixel objects of different nature, leading to more frequent erroneous results. Serious errors can result because the geometric mean is strongly affected by the discrete samples. Therefore, the good classification denoising effect cannot be obtained under the two conditions of too large or too small N value. To address these issues, the basic classification method is supplemented:
1) soft classification;
2) virtual multi-visualization processing;
3) and (4) multi-resolution processing.
These three problems will be briefly explained below:
soft classification
The soft classification is the core of the denoising algorithm proposed by us, and the two denoising algorithms with complementary characteristics in homogeneous and heterogeneous regions of an image are effectively combined. The characteristics of the denoising algorithm depend on the properties of the image, and therefore, we select the parameters most representative of the properties of the image to adjust the weighting coefficients. Instead of using hard thresholding, we compute the weighting coefficients by a smooth nonlinear logistic function, which takes values between 0, 1 as the image features change.
The result of the above equation is used as a combined weight for both algorithms. The parameters of the logistic function, i.e. the mid-point lambda, can be conveniently selected based on known statistical sampling of homogeneous and heterogeneous regions0And slope a to ensure the desired selectivity and smoothness.
Virtual multi-visualization processing
Increasing the size of the window for calculating the a/G ratio may reduce the variance of the estimated values, but may result in estimation errors. This is because the pixels entering the window have different properties after the window is enlarged. To solve this problem, we inspire a non-local denoising approach. To include more pixels, a relatively large search area is used to select those pixels that are more likely to coincide with the target pixel using the block similarity metric. Specifically, we find the K sub-blocks that are most similar to the sub-blocks to be processed using the block similarity metric over a large search range. The K sub-blocks form a three-dimensional stack. And then averaging the three-dimensional stack along the third dimension, and calculating the A/G statistics from the obtained average sub-blocks. This amounts to the use of a multi-visualization step in terms of improving data quality, which, although a relatively crude strategy, is a significant improvement in reliability.
Multi-resolution processing
The quality of the weight mapping graph is improved by establishing the three-dimensional stack in the step (1-2-1) similar to virtual visualization processing. Reliable statistics are provided even with a relatively small estimation window, thereby ensuring high resolution. However, using a fixed size sub-block window results in discontinuities at the window boundaries. This phenomenon is due to the fact that each pixel belongs to the estimation window of its many neighbors, affecting their heterogeneity index. For example, strong noise on a uniform background will cause pixels within the entire window to be marked as heterogeneous regions. Finally, the pixels of the neighborhood that should belong to a homogeneous region are denoised as a heterogeneous region, producing artifacts in the output. To solve this problem, we use a simple heuristic strategy of computing weights based on sub-block size, which may include larger weights and smaller weights, and then averaging the sub-blocks of different sizes. This can reduce some unreasonable estimates and avoid the occurrence of artifacts from the artificial filtering.
The second step comprises the following specific steps:
step (2-1): two currently most advanced denoising tools, LPG-PCA and BM3D, were chosen.
Step (2-2): the LPG-PCA algorithm has strong noise suppression capability in a homogeneous region of an image, so that a result obtained after the LPG-PCA is denoised in the homogeneous region is endowed with a large weight.
Step (2-3): the BM3D can well protect the details and the textures of the image, so that the result of denoising the BM3D in a heterogeneous region of the image is endowed with a larger weight.
The third step comprises the following specific steps:
step (3-1): and (3) determining the weight of each sub-block according to the image weight mapping graph obtained in the step (1), and using the weight to linearly combine the outputs of the two filters.
Step (3-2): according to the step (2), denoising the noise image by adopting BM3D and LPG-PCA respectively to obtain two output images;
step (3-3): and (3) effectively fusing the two output images obtained in the step (2) by using the calculated weight value graph, wherein the weights of the sub-blocks corresponding to the LPG-PCA in the homogeneous region after being denoised are larger, and the weights of the corresponding sub-blocks in the heterogeneous region after being denoised are larger in BM 3D. And effectively fusing the denoised images by the two methods to obtain the final denoised image.
Although the denoising effect of the NLM algorithm is relatively good, the structural information of the original image cannot be sufficiently protected, the image processed by the BM3D algorithm has stronger similarity between image blocks, the performance of the NLM algorithm is further improved, the image details are well protected, the signal-to-noise ratio is higher, and the visual effect is better; the LPG-PCA has strong denoising capability in a homogeneous region, a denoised image is clear, an image presented after the K-SVD processing is fuzzy, and the detail protection and denoising performance of the LPG-PCA are weaker than those of an LPG-PCA algorithm.
As a reference for selection of the classification fusion algorithm, the BM3D algorithm is selected to be combined with the LPG-PCA algorithm, the NLM algorithm is selected to be combined with the K-SVD algorithm, and simulation performance is compared. Experimental results show that the BM3D algorithm and the LPG-PCA algorithm can effectively remove noise, well protect details and textures of an image, avoid artificial artifacts and achieve the purpose of image denoising.
Fig. 2(a) -fig. 2(h) are comparison diagrams of selecting different windows for the Lena image to perform soft classification. Where FIG. 2(a) is an original Lena image; FIG. 2(b) is a weight mapping diagram obtained by partitioning the original Lena image by 3 × 3; FIG. 2(c) is a weight mapping diagram obtained by 5 × 5 blocking of the original Lena image; FIG. 2(d) is a weight mapping diagram obtained by 7 × 7 blocking the original Lena image; FIG. 2(e) is a weight mapping diagram obtained by partitioning the original Lena image by 9 × 9; FIG. 2(f) is a weight mapping obtained by partitioning the original Lena image by 10 × 10; FIG. 2(g) is a weight map obtained by partitioning the original Lena image by 11 × 11; fig. 2(h) is a weight map obtained by averaging a plurality of maps. It is apparent that fig. 2(h) effectively avoids artifacts.
Fig. 3(a) -3 (h) are comparison diagrams of Lena images containing noise and denoised by different methods. Wherein, fig. 3(a) is a noise image obtained by adding gaussian noise with a variance of 10 to the original Lena image of fig. 2 (a); FIG. 3(b) is the image of FIG. 3(a) after being denoised by BM3D method; FIG. 3(c) is the image of FIG. 3(a) after being denoised by the LPG-PCA method; FIG. 3(d) is a fused image denoised by BM3D and LPG-PCA method for FIG. 3 (a); FIG. 3(e) is a weight map after multi-view averaging; FIG. 3(f) is the image of FIG. 3(a) after being denoised by NLM method; FIG. 3(g) is the image of FIG. 3(a) after denoising by the K-SVD method; FIG. 3(h) is a fused image obtained by denoising the image of FIG. 3(a) by the NLM and K-SVD methods. Clearly, the visual effect is best in fig. 3 (d).
The above description is only a preferred embodiment of the present application and is not intended to limit the present application, and various modifications and changes may be made by those skilled in the art. Any modification, equivalent replacement, improvement and the like made within the spirit and principle of the present application shall be included in the protection scope of the present application.
Claims (5)
1. The image denoising method based on homogeneous and heterogeneous region fusion is characterized by comprising the following steps:
step (1): setting the sliding step length of a window, dividing an original image into a plurality of sub-blocks according to the set size of the window, and calculating the weight coefficient of each sub-block;
step (2): carrying out denoising treatment on all sub-blocks divided from the original image by adopting an LPG-PCA algorithm;
denoising all sub-blocks divided from the original image by adopting a three-dimensional block matching BM3D algorithm;
and (3): classifying each subblock into a homogeneous region or a heterogeneous region according to the comparison of the weight coefficient and a set threshold; correspondingly fusing the sub-blocks denoised by the two algorithms according to the region category and the weight coefficient of each sub-block to obtain a fused image, namely a final image; the homogeneous region is: the gray values of all pixel points in the area are within a set range; the heterogeneous region is: regions other than the homogeneous region.
2. The image denoising method based on homogeneous and heterogeneous region fusion as claimed in claim 1, wherein in the step (1), the step of calculating the weight coefficient of each sub-block comprises:
a step (101): calculating generalized likelihood ratio lambda of jth sub-blockj(x) Comprises the following steps:
wherein G represents the geometric mean of the jth sub-block, A represents the arithmetic mean of the jth sub-block,
a step (102): according to a generalized likelihood ratio λj(x) Calculating the weight omega (lambda) of the jth sub-blockj)：
Wherein the parameter lambda0Taking the intermediate value of the generalized likelihood ratio lambda (x) of all the sub-blocks, and taking the slope alpha as a set value.
3. The image denoising method based on homogeneous and heterogeneous region fusion as claimed in claim 1, wherein the step (3) comprises:
judging whether the jth sub-block of the image belongs to a homogeneous region or a heterogeneous region;
if the weight coefficient ω (λ)j) If the sub-block is greater than or equal to 0.5, the sub-block belongs to a heterogeneous region; for the sub-blocks belonging to the heterogeneous region, multiplying the pixel gray value of the jth sub-block subjected to denoising processing by the three-dimensional block matching BM3D algorithm by the weight omega (lambda)j) Then, multiplying the pixel gray value of the jth sub-block subjected to denoising processing by the LPG-PCA algorithm by (1-omega (lambda)j) Post-summation to obtain the pixel gray value of the jth sub-block of the fused image;
if the weight coefficient ω (λ)j) Less than 0.5, the sub-block belongs to the homogeneous region; for the sub-blocks belonging to the homogeneous region, multiplying the pixel gray value of the jth sub-block subjected to denoising processing by the three-dimensional block matching BM3D algorithm by the weight omega (lambda)j) Then, multiplying the pixel gray value of the jth sub-block subjected to denoising processing by the LPG-PCA algorithm by (1-omega (lambda)j) Post-summation to obtain the pixel gray value of the jth sub-block of the fused image;
and further obtaining pixel gray values of all sub-blocks of the fused image, namely obtaining the image of the original image after denoising.
4. Image denoising system based on homogeneity and heterogeneous region fusion, characterized by includes: a memory, a processor, and computer instructions stored on the memory and executed on the processor, the computer instructions, when executed by the processor, performing the steps of the method of any of claims 1-3.
5. A computer-readable storage medium having stored thereon computer instructions which, when executed by a processor, perform the steps of the method of any one of claims 1 to 3.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201811061739.8A CN109242797B (en) | 2018-09-12 | 2018-09-12 | Image denoising method, system and medium based on homogeneous and heterogeneous region fusion |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201811061739.8A CN109242797B (en) | 2018-09-12 | 2018-09-12 | Image denoising method, system and medium based on homogeneous and heterogeneous region fusion |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN109242797A CN109242797A (en) | 2019-01-18 |
| CN109242797B true CN109242797B (en) | 2021-10-19 |
Family
ID=65067611
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201811061739.8A Expired - Fee Related CN109242797B (en) | 2018-09-12 | 2018-09-12 | Image denoising method, system and medium based on homogeneous and heterogeneous region fusion |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN109242797B (en) |
Families Citing this family (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111612706B (en) * | 2020-04-28 | 2023-10-13 | 广州科易光电技术有限公司 | Filtering method and system applied to infrared image |
| TWI789071B (en) * | 2021-10-25 | 2023-01-01 | 瑞昱半導體股份有限公司 | Image processing system and related image processing method for image enhancement based on region control and texture synthesis |
Citations (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN103093434A (en) * | 2013-01-27 | 2013-05-08 | 西安电子科技大学 | Non-local wiener filtering image denoising method based on singular value decomposition |
Family Cites Families (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20120281923A1 (en) * | 2011-05-02 | 2012-11-08 | Yeda Research And Development Co., Ltd. | Device, system, and method of image processing utilizing non-uniform image patch recurrence |
-
2018
- 2018-09-12 CN CN201811061739.8A patent/CN109242797B/en not_active Expired - Fee Related
Patent Citations (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN103093434A (en) * | 2013-01-27 | 2013-05-08 | 西安电子科技大学 | Non-local wiener filtering image denoising method based on singular value decomposition |
Non-Patent Citations (2)
| Title |
|---|
| 《Iterative weighted maximum likelihood denoising with probabilistic patch-based weights》;Charles-Alban Deledalle等;《 IEEE Transactions on Image Processing》;20090807;第18卷(第12期);全文 * |
| 《基于广义似然比的小波域SAR图像相干斑抑制算法》;侯建华 等;《中南民族大学学报(自然科学版)》;20150331;第34卷(第1期);全文 * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN109242797A (en) | 2019-01-18 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN112132959B (en) | Digital rock core image processing method and device, computer equipment and storage medium | |
| CN105335947A (en) | Image de-noising method and image de-noising apparatus | |
| CN111783583B (en) | SAR image speckle suppression method based on non-local mean algorithm | |
| CN109919870A (en) | A kind of SAR image speckle suppression method based on BM3D | |
| CN109509163B (en) | FGF-based multi-focus image fusion method and system | |
| CN107169962B (en) | Gray level image fast segmentation method based on space density constraint kernel fuzzy clustering | |
| CN110443775B (en) | Discrete wavelet transform domain multi-focus image fusion method based on convolutional neural network | |
| CN110348459B (en) | Sonar image fractal feature extraction method based on multi-scale rapid carpet covering method | |
| CN110349112B (en) | Two-stage image denoising method based on self-adaptive singular value threshold | |
| CN114240797B (en) | OCT image denoising method, device, equipment and medium | |
| CN104657951A (en) | Multiplicative noise removal method for image | |
| Nair et al. | Direction based adaptive weighted switching median filter for removing high density impulse noise | |
| CN101944230A (en) | Multi-scale-based natural image non-local mean noise reduction method | |
| CN112488934B (en) | CS-TCGAN-based finger vein image denoising method | |
| CN109242797B (en) | Image denoising method, system and medium based on homogeneous and heterogeneous region fusion | |
| CN108305268B (en) | Image segmentation method and device | |
| Bhardwaj et al. | A Novel Method for Despeckling of Ultrasound Images Using Cellular Automata-Based Despeckling Filter | |
| CN115082336A (en) | SAR image speckle suppression method based on machine learning | |
| JP2019125203A (en) | Target recognition device, target recognition method, program and convolution neural network | |
| Bindal et al. | Novel three stage range sensitive filter for denoising high density salt & pepper noise | |
| Md. Taha et al. | Reduction of salt-and-pepper noise from digital grayscale image by using recursive switching adaptive median filter | |
| CN114240990B (en) | SAR image point target segmentation method | |
| Sen et al. | A comparative analysis of the algorithms for de-noising images contaminated with impulse noise | |
| Suneetha et al. | An Improved Denoising of Medical Images Based on Hybrid Filter Approach and Assess Quality Metrics | |
| Zhang et al. | Multi-resolution depth image restoration |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant | ||
| CF01 | Termination of patent right due to non-payment of annual fee |
Granted publication date: 20211019 |